Python API Reference¶

skill-reach provides a programmatic Python API for embedding reachability measurement, manifest linting, and automated description optimization into custom tools or workflows.

Module Index¶

Module	Description
`reach.artifact`	Evaluation artifact schema, scoring metrics, and confusion matrix data structures.
`reach.catalog`	Catalog discovery, filesystem scanning, and YAML frontmatter parsing.
`reach.check`	Two-stage CI/CD quality gate orchestration, threshold enforcement, and GitHub Actions summary generation.
`reach.cluster`	Modularity-based skill clustering for subagent catalog scoping.
`reach.config`	Configuration loading and Pydantic settings models.
`reach.diff`	A/B evaluation comparison, noise floor estimation, and effect size reporting.
`reach.lint`	Static analysis engine for validating skill manifests and listing budgets.
`reach.metrics`	Accuracy, precision, recall, F1, and multi-step trajectory evaluation metrics.
`reach.models`	Domain representations: `Skill`, `Catalog`, `Query`, and `ProbeResult`.
`reach.optimize`	Synthesis and empirical optimization routines for skill descriptions.
`reach.overlap`	BM25 vocabulary competition ranking, nearest-rival extraction, and pairwise collision detection.
`reach.queries`	Labeled evaluation query set loading, serialization, provenance recording, and digest generation.
`reach.registry`	Google Cloud Agent Registry REST client, ADC token resolution, and cache manager.
`reach.retrieval`	BM25 lexical scoring, dense embeddings, and reciprocal rank fusion (RRF).
`reach.run`	Execution engine for conducting evaluation runs and probe batches.
`reach.runtime`	Agent execution runtime interfaces, CLI subprocess template drivers, Antigravity domain bridges, and trajectory telemetry.
`reach.sweep`	Multi-scale catalog scaling sweeps, capacity knee detection, and loss decomposition.
`reach.uncertainty`	Wilson score confidence intervals, sample size sizing, and power analysis.

Quick Example¶

Programmatically load a catalog, run static linting, and inspect issues:

from pathlib import Path
from reach.catalog import load_skills
from reach.lint import lint_skills

skills = load_skills(Path("./skills"))
report = lint_skills(skills)

for issue in report.issues:
    print(f"[{issue.rule}] {issue.skill}: {issue.message}")